Multiple spectrometers with the same structure and their calibration methods

Abstract

The present invention relates to a method for calibrating a plurality of spectrometers having the same structure for component analysis. First, a plurality of samples are provided and the concentration of components in each sample is measured using a reference measurement method. The spectra of each sample are measured using a spectrometer exemplarily selected from the spectrometers having the same structure in order to determine a provisional regression model. At least one spectrum is selected from the measured spectra and / or the average spectrum formed from the measured spectra. A plurality of error spectra (01) are generated using a mathematical model of the spectrometers having the same structure. Each error spectrum (01) is added to each selected spectrum respectively in order to obtain simulated spectra. Thus, a final regression model is determined, in which the amplitudes of the measured spectra and the amplitude values of the selected simulated spectra form independent variables, and in which the reference measurement values form dependent variables. In addition, the present invention relates to a plurality of spectrometers having the same structure for component analysis.

Description

Technical Field

[0001] The present invention firstly relates to a method for calibrating a plurality of structurally identical spectrometers for component analysis. Through calibration, it should be possible to accurately measure the concentration of a specific component of a sample to be measured using each spectrometer. The calibration process at least includes providing a regression model that is used to determine the concentration of the component based on the spectrum of the sample measured by the spectrometer. In addition, the present invention also relates to a plurality of structurally identical spectrometers for component analysis. Background Art

[0002] DE 601 14 036T2 shows a method for characterizing a spectrometer instrument according to instrument variations existing between instruments or according to variations over time within the same instrument. A plurality of spectra of known standards are set by at least one spectrometer instrument. Based on spectral features extracted from at least one spectrum, at least one spectrometer instrument is classified into at least one of a plurality of predefined clusters. At least one calibration model is set for each predefined cluster. Each calibration model compensates for the instrument variations of the instrument classified into the respective cluster.

[0003] In the literature “A Review of Calibration Transfer Practices and Instrument Differences in Spectroscopy” by Workman JR, J.J., Applied Spectroscopy, Vol. 72(3), 2018, pp. 340 - 365, DOI: 10.1177 / 0003702817736064, an overview of methods for transferring calibration models between spectrometers is provided.

[0004] DE 696 08 252T2 shows a method for standardizing a plurality of spectrometers.

[0005] A method for wavelength calibration of a spectrometer is known from EP 1 998 155 A1, which is based on the principle of stepwise relative movement of corresponding measurement value blocks of a model and a calibration spectrum.

[0006] DE 101 52 679A1 teaches a method for fully automatic transfer of calibration for an optical emission spectrometer.

[0007] In DE 10 2004 061 178 A1, a method for fully automatic transfer of calibration between spectrometers is shown. The spectrometer includes a spectrometer optical element with a positionable slit.

[0008] DE 692 11 163 T2 shows a method for calibrating a spectrometer for determining characteristic values of an unknown sample by referring to a spectrometer having a first calibration equation and using constants in a second calibration equation for the instrument. For each link corresponding to a set of calibration transfer standards, based on first spectral data for each link of the set of calibration transfer standards, a value of a first dependent variable is determined from the first calibration equation. A second constant in the second calibration equation is determined such that the sum of the absolute differences between each mentioned value of a second independent variable for each link of the set of calibration transfer standards in the case of using second spectral data measured by the instrument and the corresponding values of the first dependent variable determined by the first calibration equation for each link of the set of calibration transfer standards is minimized. The mentioned characteristic values in the unknown sample are determined by using the instrument.

[0009] Since reference measurement methods are used to determine the respective component concentrations, the calibration of spectrometers for component analysis is costly because reference measurement methods mostly require chemical analysis. Therefore, it is desirable not to have to calibrate each individual finished product of a mass-produced spectrometer type for component analysis. For this purpose, the exemplary implemented calibration is designed to be stable with respect to small differences in the manufacture of the spectrometers. Exactly in the context of the increasing installation of micro spectrometers as sensors, this problem is becoming increasingly important for manufacturers and calibrators. Among a large number of spectrometers used as measuring sensors, not every product on each individual device can be compared with respect to the reference measurement methods to be carried out in the laboratory. According to the prior art, more spectrometers are used for calibration in order to minimize the influence of individual devices. In order to achieve good inter-instrument differences (IIA), spectrometers must be selected for the calibration work, which is associated with corresponding costs. However, in process technology applications (where samples are taken from the process after recording the spectra), this is almost impossible. In addition, in view of the improvement of IIA, it is not possible to optimize the calibration retrospectively because new samples and a large number of spectrometers are required for this. Summary of the Invention

[0010] Starting from the prior art, the object of the present invention is to be able to improve the inter-instrument differences (IIA) of spectrometers for component analysis with less effort.

[0011] This object is achieved by a method according to the invention and by a plurality of spectrometers having the same structure according to the invention.

[0012] The method according to the invention is used to calibrate a plurality of spectrometers having the same structure. The spectrometers having the same structure are used for component analysis. Calibration is carried out with respect to component analysis, i.e., determining the component concentration of a sample from the spectrum measured on the sample. The spectrometers are identical in structure at least in terms of measurement signal acquisition and measurement signal processing, wherein the spectrometers, for example, the design of the housing or the connector, may be different with respect to those characteristics that do not affect measurement signal acquisition and measurement signal processing. The spectrometers having the same structure are preferably of the same type.

[0013] The spectrometers are, for example, near-infrared spectrometers, VIS / NIR spectrometers or full-range spectrometers. The spectrometers may have a transmission structure, a transmission-reflection structure or a reflection structure. The spectrometers may be, for example, monochromator-based spectrometers, interferometer-based spectrometers, filter-based spectrometers or MEMS spectrometers.

[0014] The calibration process at least includes providing a regression model for determining the component concentration based on the spectrum measured by the spectrometer for the sample. In this regard, the method is at least used to provide a regression model for calibrating a plurality of spectrometers having the same structure for component analysis.

[0015] In one step of the method, a large number of samples are provided. The samples contain different concentrations of the component. Here, it refers to the component to be measured for spectrometer calibration. Preferably, at least ten samples are provided, and more preferably at least 100 samples are provided. Preferably, 50 to 500 samples are provided. The samples are taken from the material or product to be inspected by the spectrometer in terms of its content. The material or product is preferably an agricultural product, food or foodstuff. The agricultural product is preferably grain. The spectrometer is preferably configured to measure the concentration of at least one component of the agricultural product or foodstuff on an agricultural vehicle, in agricultural machinery or in a food manufacturing process. The component is preferably formed by water, protein, oil, sugar, starch or crude fiber. The measurement of the water concentration is moisture measurement.

[0016] In a further step of the method, the component concentration in each sample is measured using a reference measurement method, thereby obtaining a reference measurement value of the component concentration. The reference measurement method obtains an accurate measurement value of the component concentration in each sample. The reference measurement method is preferably formed by a chemical analysis method. The chemical analysis method is preferably a wet chemical analysis method.

[0017] In a further step of the method, the spectrum of each sample is measured using a spectrometer exemplarily selected from the spectrometers having the same structure. The spectrum is preferably an absorption spectrum, a transmission spectrum and / or a reflection spectrum. The spectra are measured in the samples respectively according to the same measurement principle.

[0018] The spectrum is preferably measured in the infrared range, in the visible range and / or in the ultraviolet range of the electromagnetic spectrum. The spectrum is preferably measured in the near infrared range of the electromagnetic spectrum. The spectrum is measured over a wavelength range of preferably at least 100 nm, and further preferably at least 300 nm in size. The spectrum is preferably measured by measuring the amplitude at selected wavelengths. Preferably, the amplitude is measured at at least 10 selected wavelengths. Further preferably, the amplitude is measured at at least 100 selected wavelengths. Further preferably, the amplitude is measured at at least 300 selected wavelengths.

[0019] The spectrometer selected by way of example is of identical construction to other spectrometers of identical construction. The spectrometer selected by way of example preferably has an average error.

[0020] In a further step of the method, a provisional regression model is determined in which the amplitude values ​​of the measured spectrum form the independent variable and in which the reference measured values ​​of the component concentration form the dependent variable. Thus, the following mathematical function can be used: which describes the correlation of the component concentration with the amplitude values ​​of the spectrum. The provisional regression model preferably has at least one further dependent variable describing a physical or chemical property of the sample.

[0021] In a further step of the method, at least one spectrum is selected from a number of spectra, wherein the number of spectra includes measured spectra and average spectra formed from the measured spectra. The selection is based on suitability for predicting component concentrations, in particular in the respective sample. Preferably, one to ten spectra are selected from the measured spectra and / or average spectra formed from the measured spectra.

[0022] The step of selecting at least one spectrum from the measured spectrum and / or the average spectrum formed by the measured spectrum preferably includes one or both of the two sub-steps described below. According to one sub-step, a temporary regression model is applied to the measured spectrum in order to obtain the predicted values ​​of the component concentration in the respective samples, respectively, wherein the following measured spectrum or multiple measured spectra are selected: for this spectrum or these spectra, the predicted value is closest to the reference measured value of the respective sample. Therefore, the following measured spectrum is selected: the spectrum is as close as possible to the regression line of the temporary regression model. According to another sub-step, the average spectrum is selected. The average spectrum is preferably formed by the measured spectrum so that the amplitude values ​​of the measured spectrum are averaged at each selected wavelength, for example by arithmetic averaging. The reference measured value of the component concentration is assigned to the average spectrum, and the reference measured value is determined by applying the temporary regression model to the average spectrum.

[0023] In a further step of the method, the tolerances of the components of spectrometers with the same structure are simulated multiple times using a mathematical model of a spectrometer with the same structure in order to obtain a plurality of error spectra. The mathematical model of the spectrometer with the same structure describes the components of the spectrometer with the same structure; in particular, the optical and electronic components. The mathematical model particularly describes the error characteristics, i.e., the tolerances of the components that occur, such that the mathematical model is suitable for calculating the maximum error in the correctness for each wavelength and the maximum error in the correctness for the respective amplitudes of transmission, reflection, and / or absorption. Based on the maximum errors for the respective selected wavelengths, possible error spectra are simulated relative to an ideal spectrometer. The error spectra represent difference spectra. Preferably, at least 100 simulations are performed in order to obtain at least 100 error spectra. More preferably, at least 500 simulations are performed in order to obtain at least 500 error spectra.

[0024] In a further step of the method, the respective error spectra are added to the selected spectra or the respective selected spectra separately in order to obtain simulated spectra. The simulated spectra are spectra that could be actually measured using a spectrometer with the same structure because, on the one hand, they are based on spectra actually measured using a spectrometer with the same structure and, on the other hand, have an error component that is known by simulation using the mathematical model of the spectrometer with the same structure. Preferably, the respective error spectra are added to the selected spectra or the respective selected spectra separately for the amplitude values of the selected wavelengths. When the number of error spectra is, for example, 1000 and the number of selected spectra is, for example, one, the number of simulated spectra obtained is 1×1000 = 1000. When the number of error spectra is, for example, 1000 and the number of selected spectra is, for example, 5, the number of simulated spectra obtained is 5×1000 = 5000.

[0025] In a further step of the method, a provisional regression model is applied to the simulated spectra in order to obtain predicted values of the component concentrations separately. Thus, for each simulated spectrum, there is a predicted value of the component concentration. A reference measurement value of the component concentration is assigned to the simulated spectra separately, and this reference measurement value is known for the selected spectra based on the respective simulated spectra.

[0026] In a further step of the method, a certain number of simulated spectra are selected, where the predicted values obtained for the selected simulated spectra reflect the deviation of the predicted values obtained for the simulated spectra. Thus, those simulated spectra that are suitable for improving the provisional regression model are selected. Preferably, the number of selected simulated spectra is between 50 and 1000.

[0027] In a further step of the method, a final regression model is determined, in which the amplitude values of the measured spectra and the amplitude values of the selected simulated spectra form the independent variables, and in which the reference measured values of the component concentrations form the dependent variable. The determination of the final regression model differs from the determination of the provisional regression model only in that the amplitude values of the selected simulated spectra are additionally taken into account. Thus, the final regression model is an improvement over the provisional regression model, so that the final regression model represents a more reliable calibration against errors, in particular against differences caused by the manufacturing tolerances of spectrometers of the same construction.

[0028] The independent variables of the provisional regression model and the independent variables of the final regression model are preferably the amplitude values of the measured or selected simulated spectra at the selected wavelengths, respectively. The independent variables of the provisional regression model and the independent variables of the final regression model are preferably the amplitude values of the measured or selected simulated spectra of the same selected wavelengths, respectively. The provisional regression model and the final regression model are preferably formed by a multiple linear regression model, respectively. Preferably, the determination of the provisional regression model and the determination of the final regression model are each achieved by partial least squares regression.

[0029] Preferably, the number of selected simulated spectra is at most 50% of the number of measured spectra. For example, the number of selected simulated spectra is about 75, while the number of measured spectra is about 150. This achieves a significantly higher reliability of the final regression model compared to the provisional regression model.

[0030] A particular advantage of the described method is that the calibration of spectrometers of the same construction for component analysis can be improved by less costly simulations. It is not necessary to measure with each spectrometer of the same construction to individually calibrate these spectrometers for component analysis. By using a mathematical model, the error characteristics of spectrometers of the same construction are taken into account. The discrete width of spectrometers of the same construction is reduced by up to 30% when using the same model and without further adjustment depending on the product and the component.

[0031] In a preferred embodiment, the method is also configured to utilize a calibration or final regression model. To this end, the method includes an additional step in which the final regression model is used in spectrometers of identical construction in order to determine the component concentration of a sample using the spectrometers based on the spectra measured for the sample with the respective spectrometers. To this end, the final regression model in software form is loaded into the respective spectrometers, which software describes the final regression model as a mathematical relationship between the component concentration and the measured spectrum. Alternatively, the final regression model is preferably used in a network. The network includes spectrometers of identical construction and at least one computing unit. The network is a data network. The spectrometers and the at least one computing unit are connected to one another via a data connection. The at least one computing unit is preferably formed by a computer, in particular a server. Preferably, the network includes a plurality of computers. The at least one computing unit is used to determine the component concentration of a sample based on the spectrum measured for the sample with the spectrometers. Thus, the determination of the concentration is preferably carried out by cloud computing. To this end, the final regression model in software form is transmitted to the at least one computing unit, which software describes the final regression model as a mathematical relationship between the component concentration and the measured spectrum.

[0032] According to the invention, a plurality of spectrometers of identical construction for component analysis are also disclosed. Each spectrometer is respectively configured to determine the component concentration of a sample from the spectrum measured with the respective spectrometer. The spectrometers are calibrated according to the method described above such that the relationship between the measured value to be determined for the component concentration defined by the final regression model and the measured spectrum is established. Preferably, one of the above-described preferred embodiments of the method is used to determine the final regression model.

[0033] The plurality of spectrometers of identical construction preferably includes at least 100 spectrometers of identical construction, and more preferably includes at least 1000 spectrometers of identical construction.

[0034] The spectrometers preferably also have the features described above with respect to the method. Description of the Drawings

[0035] Further details and refinements of the invention result from the following description of the preferred embodiments of the invention with reference to the accompanying drawings. Among them:

[0036] Figure 1 A diagram showing the error spectrum produced in accordance with a preferred embodiment of the method according to the invention is shown; and

[0037] Figure 2 A diagram showing the predicted values determined according to the prior art and in accordance with a preferred embodiment of the method according to the invention is shown. Detailed Description of the Invention

[0038] Figure 1 A graph is shown that depicts an error spectrum 01 generated according to a preferred embodiment of the method according to the present invention. The error spectrum 01 is generated in the following manner, namely, by using a mathematical model to simulate the tolerances of components of structurally identical spectrometers to be calibrated for component analysis, wherein the mathematical model depicts one of a plurality of structurally identical spectrometers. The x-axis of the graph represents the wavelength λ. The deviation, i.e., the error generated compared to a predetermined value of the wavelength in the respective application of the mathematical model, is plotted on the y-axis of the graph. An ideal spectrometer has zero error. Also shown in the graph is the 3σ deviation 02 obtained using the mathematical model. The error spectrum 01 can be used to develop a regression model for more reliably calibrating a plurality of structurally identical spectrometers to be calibrated for component analysis against the error of the spectrometers.

[0039] Figure 2 A graph is shown that depicts a predicted value 03 determined according to the prior art and a predicted value 04 determined according to a preferred embodiment of the method according to the present invention. A plurality of spectrometers are plotted on the x-axis. The magnitude of the predicted value of the concentration of the component measured using the respective spectrometer is plotted on the y-axis. The predicted values 03, 04 are classified according to the magnitude of their deviation respectively. The predicted value 04 determined according to a preferred embodiment of the method according to the present invention shows that the between-instrument difference (IIA) is significantly improved compared to the prior art.

[0040] List of reference numerals

[0041] 01 Error spectrum

[0042] 02 3σ deviation

[0043] 03 Predicted value according to the prior art

[0044] 04 Predicted value according to the present invention

Claims

1. A method for providing a regression model to calibrate multiple spectrometers with the same structure for component analysis, the method comprising the following steps: - Providing multiple samples containing components with different concentrations; - Measuring the component concentrations in each sample using a reference measurement method to obtain reference measurement values of the component concentrations; - Measuring the spectra of each sample using a spectrometer exemplarily selected from the spectrometers with the same structure; - Determining a temporary regression model, in which the amplitude values of the measured spectra form independent variables, and in which the reference measurement values of the component concentrations form dependent variables; - Selecting at least one spectrum from the measured spectra and / or the average spectrum formed by the measured spectra according to the applicability for predicting the component concentration; - Using a mathematical model of the spectrometers with the same structure to simulate the tolerances of the components of the spectrometers with the same structure multiple times to obtain multiple error spectra (01); - Adding each error spectrum (01) to the selected spectrum or to each selected spectrum respectively to obtain simulated spectra; - Applying the temporary regression model to the simulated spectra to respectively obtain predicted values of the component concentrations; - Selecting a certain number of simulated spectra, wherein the predicted values obtained for the selected simulated spectra reflect the deviation of the predicted values obtained for the simulated spectra; And - Determining a final regression model, in which the amplitude values of the measured spectra and the amplitude values of the selected simulated spectra form independent variables, and in which the reference measurement values of the component concentrations form dependent variables.

2. The method according to claim 1, characterized in that, The reference measurement method is formed by a chemical analysis method.

3. The method according to claim 1, characterized in that The components are formed by water, protein, oil, sugar, starch or crude fiber.

4. The method according to any one of claims 1 to 3, characterized in that, The spectra are measured in the infrared range, visible range and / or ultraviolet range of the electromagnetic spectrum.

5. The method according to any one of claims 1 to 3, characterized in that Using a spectrometer exemplarily selected from the spectrometers with the same structure, measuring the spectra of each sample in a wavelength range of at least 300 nm, wherein the spectra are measured by measuring the amplitudes at at least 10 selected wavelengths.

6. The method according to claim 5, wherein The independent variables of the temporary regression model are respectively formed by the amplitude values of the spectra measured at the selected wavelengths, and the independent variables of the final regression model are respectively formed by the amplitude values of the selected simulated spectra at the selected wavelengths.

7. The method according to any one of claims 1 to 3, characterized in that The independent variables of the temporary regression model and the independent variables of the final regression model are formed by the amplitude values of the spectra at the selected wavelengths.

8. The method according to any one of claims 1 to 3, characterized in that The temporary regression model and the final regression model are respectively formed by a multiple linear regression model.

9. The method according to any one of claims 1 to 3, characterized in that, The step of selecting at least one spectrum from the measured spectra and / or the average spectrum formed by the measured spectra includes one or two sub-steps of the following two sub-steps: - Applying the temporary regression model to the measured spectra to respectively obtain predicted values of the component concentrations in their respective samples, wherein one or more of the following measured spectra are selected: for the one or more measured spectra, the predicted values are closest to the reference measurement values of their respective samples; - Selecting the average spectrum.

10. The method according to any one of claims 1 to 3, characterized in that, Select one to ten spectra from the measured spectra and / or the average spectrum formed from the measured spectra.

11. The method according to any one of claims 1 to 3, characterized in that, Perform at least 100 simulations using a mathematical model of spectrometers with the same structure to simulate the tolerances of components of spectrometers with the same structure, thereby obtaining at least 100 error spectra (01).

12. The method according to any one of claims 1 to 3, characterized in that, The number of selected simulated spectra is between 50 and 1000.

13. The method according to any one of claims 1 to 3, characterized in that, The method includes the following additional steps: - Use the final regression model in each spectrometer with the same structure to determine the component concentration of a sample by the spectrometer based on the spectrum of the sample measured by the respective spectrometer.

14. The method according to any one of claims 1 to 3, characterized in that The method includes the following additional steps: - Use the final regression model in a network including each spectrometer with the same structure and at least one computing unit, wherein the at least one computer unit is configured to determine the component concentration of a sample based on the spectrum of the sample measured by the spectrometer.

15. A plurality of spectrometers with the same structure for component analysis, each spectrometer being configured to determine the component concentration of a sample from the spectrum measured by the respective spectrometer, and the spectrometers being calibrated using the method according to any one of claims 1 to 14 such that the relationship between the measured value of the component concentration to be determined defined by the final regression model and the measured spectrum is established.

Images